Advanced
Open
Pro
Decide whether a conversion-rate drop is an anomaly or a changepoint
Part of the ML System Design Interview path →
Part of the Advanced Data Science & Experimentation path →
A checkout conversion rate has run at about 3.1 percent with a weekly pattern for a year. Your detector flags a drop to 2.6 percent on Tuesday. Two scenarios follow: (a) it goes back to 3.1 percent on Thursday; (b) it stays at about 2.6 percent for three weeks.
- How do an anomaly detector and a changepoint detector see each scenario?
- Which algorithms would you use for each, and with what settings?
- What should happen operationally in (a) and (b)? What mistakes happen if you treat (b) as (a), or the reverse?
Share this question